trustyai.explainers.LimeExplainer

class trustyai.explainers.LimeExplainer(**kwargs)

“Which features were most important to the results?”

LIME (Local Interpretable Model-agnostic Explanations) seeks to answer this question via providing saliencies, weights associated with each input feature that describe how strongly said feature contributed to the model’s output.

__init__(**kwargs)

Initialize the LimeExplainer.

Parameters:
Keyword Arguments:
  • penalise_sparse_balancebool

    (default= True) Whether to penalise features that are likely to produce linearly inseparable outputs. This can improve the efficacy and interpretability of the outputted saliencies.

  • normalise_weightsbool

    (default= False) Whether to normalise the saliencies generated by LIME. If selected, saliencies will be normalized between 0 and 1.

  • use_wlr_modelbool

    (default= True) Whether to use a weighted linear regression as the LIME explanatory model. If false, a multilayer perceptron is used, which generally has a slower runtime,

  • seed: int

    (default= 0) The random seed to be used.

  • perturbations: int

    (default= 1) The starting number of feature perturbations within the explanation process.

  • trackCounterfactualsbool

    (default= False) Keep track of produced byproduct counterfactuals during LIME run.

  • samples: int

    (default= 300) Number of samples to be generated for the local linear model training.

  • encoding_params: Union[list, tuple]

    (default= (0.07, 0.3)) Lime encoding parameters, as a tuple/list of two float numbers: - encoding_params[0] is the width of the Gaussian filter for clustering number features. - encoding_params[1] is the threshold for clustering number features.

  • data_distribution: PredictionInputsDataDistribution

    (default= PredictionInputsDataDistribution([])) Data distribution used to find better feature perturbations

  • features: int

    (default= 6) Number of feature to select from the original set of input features

  • retries: int

    (default= 3) Number of retries performed by LIME to find a separable dataset

  • dataset_minimum: int

    (default= 10) Minimum number of samples retained by the proximity filter to be acceptable

  • separable_dataset_ratio: float

    (default= 0.1) Minimum portion of the encoded dataset that needs to have a different label

  • kernel_width: float

    (default= 0.5) Width of the proximity kernel

  • proximity_threshold: float

    (default= 0.83) Proximity threshold used to retain close samples

  • adapt_dataset_variance: bool

    (default= True) Whether LIME should try to increase the perturbation variance in subsequent retries

  • feature_selection: bool

    (default= True) Whether LIME should generate saliency for to the most important features only

  • filter_interpretable: bool

    (default= False) Whether the proximity filter should happen in the interpretable space

Methods

__init__(**kwargs)

Initialize the LimeExplainer.

explain(inputs, outputs, model)

Produce a LIME explanation.
explain(inputs: int | float | integer | inexact | ndarray | DataFrame | Series | List[Feature] | PredictionInput, outputs: int | float | integer | inexact | ndarray | DataFrame | Series | List[Output] | PredictionOutput, model: PredictionProvider | Model) → LimeResults

Produce a LIME explanation.

Parameters:
inputsint, float, numpy.number, List[Union[int, float, numpy.number]], numpy.ndarray, pandas.DataFrame, pandas.Series, List[Feature], or PredictionInput

The input features to the model, as a:

  • If there’s only a single input feature, an int, float, or any of the Numpy equivalents can be used.

  • A list of int, float, or any of the Numpy equivalents.

  • Numpy array of shape [1, n_features] or [n_features]

  • Pandas DataFrame with 1 row and n_features columns

  • Pandas Series with n_features rows

  • A List of TrustyAI Feature, as created by the feature() function

  • A TrustyAI PredictionInput

outputsint, float, numpy.number, List[Union[int, float, numpy.number]], numpy.ndarray, pandas.DataFrame, pandas.Series, List[Output], or PredictionOutput

The corresponding model outputs for the provided features, that is, outputs = model(input_features). These can take the form of a:

  • If there’s only a single output, an int, float, or any of the Numpy equivalents can be used.

  • A list of int, float, or any of the Numpy equivalents.

  • Numpy array of shape [1, n_outputs] or [n_outputs]

  • Pandas DataFrame with 1 row and n_outputs columns

  • Pandas Series with n_outputs rows

  • A List of TrustyAI Output, as created by the output() function

  • A TrustyAI PredictionOutput

modelPredictionProvider

The TrustyAI PredictionProvider, as generated by Model or ArrowModel.

Returns:
LimeResults

Object containing the results of the LIME explanation.